Researchers proposed two randomized approximations for spectral co-clustering of word-document matrices, reducing computational cost on high-dimensional data. The approximations use random projection or element-wise sampling and show varying performance depending on matrix sparsity.